Classification and linguistic characterization of non-deterministic data

Sequential classification is here a problem of estimation of a fuzzy partition. Parameters of membership functions can be estimated recursively by a learning or selflearning mechanism. Classes can be described by multivariable possibility distributions. The marginal possibility distribution for one of the components describing an object can be matched to a possibility distribution defined in [0,1] chosen among a set corresponding to linguistic labels as 'low', 'high', 'very high', ..., and a label can be associated with this component. A higher level learning may be introduced to modify the semantics of the linguistic labels.